Re: circular markersizes via rotation matrix

[Daniel J Sebald]

On 08/19/2016 04:22 PM, Rik wrote:
> 8/19/16
>
> Dan,
>
> I benchmarked the current code for creating circular markers ("o") using
> either the current method relying on trigonometry and sin()/cos() function
> calls or via the rotation matrix that you coded which only uses
> multiplications and additions.  The short story is that I didn't find any
> significant difference so I left the code as is.

That's similar to what I had found.  I ran your scripts here and saw no 
difference as well.  The issue, I think, is that the rendering routine 
for the circles is a tiny fraction of the cpu consumed elsewhere.  My 
guess is that the creation of the properties associated with each new 
"plot()" call consumes a lot of CPU.  There is clearing old memory, 
mallocing new memory, initializing all the properties, etc.  The OpenGL 
code, too, is probably considerable.

I think if we want to test the efficiency of using trig functions all 
the way around the circle versus using matrix rotations, we'd probably 
have to write a very simple C++ test routine and basically just do the 
computations for, say, 5000000 circles and see which versions runs 
faster.  One doesn't even have to call any OpenGL code for that.  It's 
the same number of points--it's only the way they are computed that is 
different.

Or, maybe another crude way is to draw 50000 circles on a plot:

graphics_toolkit ('qt')
plot(rand(50000,1),rand(50000,1), 'o');

and drag the window corner around to see how responsive the redraw is.

I actually like the C-program idea.  It wouldn't be but a twenty to 
thirty line program.  If I find time late tonight I'll do that.

Will it matter in the bigger picture?  Not much.

Dan